SEARCH

SEARCH BY CITATION

Keywords:

  • anisotropy;
  • monitoring;
  • shear wave

Abstract

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Travel Time Monitoring With ACROSS
  5. 3. Changes in S Wave Travel Times Associated With Earthquakes in the Middle Distance
  6. 4. Discussion
  7. 5. Conclusion
  8. Appendix A:: Correction of the Source Instability
  9. Acknowledgments
  10. References

[1] Temporal variation in shear wave anisotropy was detected in a monitoring experiment using an accurately controlled, routinely operated signal system (Accurately Controlled Routinely Operated Signal System, ACROSS). We conducted an experiment that lasted for 15 months between January 2000 and April 2001 at a site near the Nojima fault, which ruptured during the 1995 Kobe earthquake (Mw 7.2) [Yamaoka et al., 2001; Ikuta et al., 2002]. Two vibration sources that generated 2 × 105 N with centrifugal force, which were firmly fixed on the ground, were used to emit elastic waves. Seismometers deployed at the bottom of 800-m- and 1700-m-deep boreholes near the ACROSS sources were used to receive the signal. We extracted small temporal changes in the travel time of the P and S wave by calculating cross-spectral density among the records every hour. During the experiment, sudden delays in travel times for the S wave were observed when the 2000 western Tottori earthquake (Mw 6.6) and the 2001 Geiyo earthquake (Mw 6.4) occurred. Their epicenters were 165 and 215 km away from the site, respectively. The travel times of the S waves between the surface and the bottom of 800-m-deep borehole abruptly slowed down and gradually speeded up with each earthquake. The delay in S was about 0.4% and 0.1% of the absolute travel time for the western Tottori earthquake and the Geiyo earthquake, respectively. The delays were polarized in a direction perpendicular to the Nojima fault in both cases. This indicates that the density of the cracks parallel to the fault increased in association with the earthquakes. These cracks can be regarded as opened by the increase in pore pressure. The small changes in P wave velocities support this interpretation. An additional experiment to determine the static anisotropy revealed that there was a preferred orientation of the cracks that was enhanced with the strong shaking of earthquakes in the middle distance.

1. Introduction

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Travel Time Monitoring With ACROSS
  5. 3. Changes in S Wave Travel Times Associated With Earthquakes in the Middle Distance
  6. 4. Discussion
  7. 5. Conclusion
  8. Appendix A:: Correction of the Source Instability
  9. Acknowledgments
  10. References

[2] Many researchers have studied temporal variations in the propagation properties of seismic waves using various methods such as VP/VS, S wave splitting, and coda Q−1 measurements.

[3] The velocity of the P and S waves has been repeatedly mentioned as an important indicator of temporal variation in crustal properties such as stress field or fluid distributions in cracks. Semenov [1969] reported premonitory decrease in the VP/VS for some moderate earthquakes in the Garm region of the Tadzhik. The VP/VS decreased by 10% of the normal VP/VS. Similar observations that refer to an M 7.5 earthquake in the Hyuganada area, Japan [Terashima, 1974] and the Blue Mountain Lake earthquakes [Aggarwal et al., 1974] have been reported. The measurements were performed using small earthquakes; thus their times of origin and their locations are ambiguous. Velocity measurements for the P and S waves have been also made using artificial sources such as quarry blasts [McEvilly and Johnson, 1974], vibroseis [Karageorgi et al., 1992], and air guns [Reasenberg and Aki, 1974]. For example, McEvilly and Johnson [1974] measured the travel times of crustal P and S waves from quarry blasts in central California passing within 10 km of the hypocenters of moderate earthquakes. They did not find any changes in the VP/VS associated with earthquake as reported by Semenov [1969].

[4] Splitting of the S wave is one of the most powerful tools for detecting temporal variation in the stress field (reviewed by Crampin [1987]). An S wave, in anisotropic media, splits into two orthogonal waves with different velocities. Crampin [1978] showed that the S wave splits into two directions that are parallel and normal to the crack plane. The leading wave is polarized into crack parallel and the lagging wave into crack normal. Recently, some papers reported that changes in anisotropy have actually been detected using S wave splitting [e.g., Bokelmann and Harjes, 2000; Tadokoro and Ando, 2002; Saiga et al., 2003; Zatsepin and Crampin, 1997; Aster et al., 1990]. For example, on the Nojima fault zone that ruptured during the 1995 Hyogo-ken Nanbu (Kobe) earthquake (M 7.2), Tadokoro and Ando [2002] observed unusual direction of the leading S wave at stations on the fault zone for a period of 9–12 months after the main shock; this was parallel to the fault and was different from the direction detected in the surrounding area. The direction changed to be parallel to the directions in the surrounding area 33–45 months after the main shock. This change in the splitting observed at the station on the fault shows healing processes of ruptures in the fault zone. Saiga et al. [2003] found spatial and temporal variations in the splitting that correspond to a moderate earthquake (M 5.7) in the area with a radius of about 100 km. The temporal and spatial variation explains the change in the Coulomb failure function (ΔCFF) caused by both coseismic and postseismic fault slips of the earthquake. Bokelmann and Harjes [2000] also observed S wave splitting in records of induced earthquakes due to a hydraulic fracturing experiment at the German Continental Deep Drilling Program (KTB) borehole. They detected decrease in anisotropy associated with induced seismicity. They suggested that the temporal variation was due to tectonic stress release from seismic events due to fluid injection.

[5] These studies use natural earthquakes, the advantage of which is that the ambiguity of the origin time does not affect on the results. If we use an artificial source, for which the origin time is accurately known, the absolute delay in the S wave can be measured, enabling us to know the change in crack density.

[6] Recently, an artificial vibration source system especially for monitoring temporal variation in the velocity of P and S waves has been developed [Yamaoka et al., 2001], the Accurately Controlled Routinely Operated Signal System (ACROSS). The primary feature of the system is to generate sinusoidal waves that are accurately controlled with reference to the GPS clock. This system enables us to continuously monitor seismic wave propagation without destroying the surrounding ground.

[7] We have conducted a continuous 15-month experiment using this system between January 2000 and April 2001 at a site on Awaji Island, Japan. Ikuta et al. [2002] reported the results of this experiment and showed the potential ability of ACROSS. They also estimated the uncertainty of S wave travel time to be no more than 0.1 ms in about 1 km between the source and the receivers. During the experiment, sudden decreases in S waves velocity were observed when the 2000 western Tottori earthquake (WT; Mw 6.6) and the 2001 Geiyo earthquake (GY; Mw 6.4) occurred. This paper focuses on changes in anisotropy at the time of the WT and the GY earthquake using S wave polarizations.

2. Travel Time Monitoring With ACROSS

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Travel Time Monitoring With ACROSS
  5. 3. Changes in S Wave Travel Times Associated With Earthquakes in the Middle Distance
  6. 4. Discussion
  7. 5. Conclusion
  8. Appendix A:: Correction of the Source Instability
  9. Acknowledgments
  10. References

[8] Long-term monitoring of seismic velocity using ACROSS was carried out between January 2000 and April 2001 near the Nojima fault, which ruptured at the 1995 Kobe earthquake (Figure 1). In the experiment, ACROSS vibrators worked almost continuously without problems except for planned or unexpected power shutdowns. As the details of this experiment have already been reported by Ikuta et al. [2002], we briefly describe the experiment at the Awaji site.

image

Figure 1. (a) Location of the ACROSS site and the four earthquakes referred to in text. The ACROSS site is marked by the open star, and the epicenters are marked by the solid stars. The heavy line indicates the Nojima fault. Four strong motion accelerometers are deployed by Japan National Research Institute for Earth Science and Disaster Prevention (NIED) (open circles). (b) Location of the ACROSS vibrators (open star) and the seismometers. The solid triangles denote the 1700-m and the 800-m-deep boreholes. The enlarged map at the bottom right shows the configuration of the observation system. The thick gray lines are traces of the 1700-m and the 800-m-deep boreholes. The open circles denote borehole-type seismometers. (c) North-south vertical cross section showing the geometry of the 800-m-and 1700-m-deep boreholes. The solid circle denotes the strain meter at the bottom of the 800-m-deep borehole.

Download figure to PowerPoint

2.1. Site

[9] Two ACROSS vibrators are deployed at the ground surface (Figure 2). One is designed to produce sinusoidal waves up to 35 Hz, and the other up to 25 Hz, with a maximum force of 2.0 × 105 N. Hereafter, they are called high-frequency (HF) and low-frequency (LF) units. The ACROSS sources are designed to generate a sinusoidal force by rotating an eccentric mass around a vertical axis. The rotation generates a single force in all directions in the horizontal plane. They are firmly fixed to the granite ground in the following manner. We built a foundation of reinforced concrete in a rectangular ditch on the granite ground. The ACROSS sources are fixed in a hole in this foundation with wedges to avoid any irregular shaking due to loose coupling.

image

Figure 2. Four velocity-type seismometers deployed on the foundation (solid circles, right, left, and middle) and at the bottom of a 10-m-deep borehole (10-m borehole). Their records are used for correcting signals received at the 800-m borehole sensor. Two ACROSS sources (solid squares) are fixed firmly to the hole in the foundation of the reinforced concrete (shaded portion) with wedges.

Download figure to PowerPoint

[10] Four three-component velocity sensors with a natural frequency of 1.0 Hz were deployed around the sources to record the actual vibrations. The records were used afterward for correction. Three of them were placed on the surface of the foundation, and one was deployed at the bottom of the 10-m-deep borehole.

[11] Two deep boreholes with depths of 800 m and 1700 m were drilled near the ACROSS vibrators. They were 120 m and 180 m away from the source, respectively. The profiles of the boreholes show that all country rock is composed of granite and that there is no structural boundary. A three-component velocity sensor, accelerometer, and a three-component strain meter were installed at the bottom of the 800-m borehole. A water pressure gauge was also installed at the top of the 800-m borehole. The value given by the gauge indicates the pore pressure at the bottom of the borehole because the borehole is filled by groundwater and sealed except for at its bottom. In the 1700-m borehole, a three-component velocity sensor was installed at the bottom. Hereafter, we call the seismometers installed at the bottoms of the two boreholes the 800-m and the 1700-m borehole sensors, respectively. The 1700-m borehole sensor has a natural period at 0.22 s, and the 800-m borehole sensor has a natural period at 0.5 s for the horizontal component and 0.33 for the vertical component. In the experiment, we aimed to detect temporal variation in the signal recorded with these sensors.

2.2. Operation Parameters for ACROSS

[12] The HF and LF sources were operated simultaneously using frequency modulation to cover as wide a frequency range as possible. The HF and LF units were operated using the same modulation amplitude of 2.5 Hz and modulation period of 5 s but with a different rotational frequency centered at 13 and 19.1 Hz, respectively.

[13] In this experiment, we adopted the frequency modulation formulated as ω(t):

  • equation image

where ω0 is the central angular frequency, Δω is the modulation amplitude, ωm is the modulation angular frequency, and q−1 is the contribution ratio of the double component of the modulating frequency. This modulation provides smaller fluctuation in spectral peaks within the controllability of motors. We adopted the values 2π × 13 and 2π × 19.1 rad/s for ω0 for the LF and the HF unit, respectively, 2π × 2.1 rad/s to Δω, 2π 5 rad/s to ωm, and 7.28 to q for both the LF and the HF units. We can identify the phases arriving at the sensors within 5 s after the excitation at the source because the sources repeated the same frequency modulation with an interval of 5 s.

[14] The data were sampled at 100 Hz with a resolution of 24 bits and stacked with an interval of 100 s. The data were saved once per hour so that each data set consisted of 36 stacks. Before stacking, each unit of 100-s-long data was normalized by its maximum amplitudes to suppress the effect of incidental large external noise such as natural earthquakes.

2.3. Analysis for Temporal Variation in the Travel Time of the ACROSS Signal

[15] In the analysis, we calculated temporal variation in the travel time for the P and S waves for the data recorded at the 800-m and the 1700-m borehole sensors for two periods including the time of the WT and the GY earthquakes. We adopted the method described by Ikuta et al. [2002] to estimate temporal variation in travel time.

[16] First, we extracted the ACROSS signals from each unit of stacked data in the frequency domain. The Fourier amplitude spectra of the observed data are shown in Figure 3. As the signals generated by the ACROSS sources were precisely repeated at 5-s intervals, they were composed of sinusoids that are multiples of 1/5 Hz. The extracted signal was divided by the force generated by the source in the frequency domain. The results are regarded as a transfer function (or band-limited frequency response) between the source and the receivers. By applying inverse Fourier transformation, we can obtain the signal in the time domain, in which P, S, and some later phases were identified (Figure 4).

image

Figure 3. Two robes ranging from 10 to 17 Hz and from 15 to 23 Hz, corresponding to the LF and the HF units, respectively. The spectral peaks of each unit have a constant interval of 0.2 Hz because a frequency modulation period of 5 s is used. Note that the spectral peaks for the two sources do not overlap because the central frequency of the HF unit has shifted by 0.1 Hz away from the integer multiples of 0.2 Hz, which is the interval of the spectral peaks.

Download figure to PowerPoint

image

Figure 4. Transfer functions between the ACROSS sources and borehole sensors. Gray patches show the part to which the Hanning windows are applied to extract P or S phases.

Download figure to PowerPoint

[17] To detect small temporal variations in the series of the wave traces, we calculated cross-spectral densities (CSD) among the traces on the sections including P or S. We extracted the P and S phases by applying the Hanning window on each trace as shown in Figure 4. All the windows are 0.3 s long, which corresponds to the sections that include five peaks and dips. The P phases for the record of the 800-m borehole sensor were extracted from the wave traces between 0.1 and 0.4 s. The S phases for the record of the 800-m borehole sensor were between 0.3 and 0.6 s. The P and S phases for the record of the 1700-m borehole sensor were between 0.25–0.55 and 0.6–0.9 s, respectively. The CSD for the WT earthquake was calculated with reference to the trace at 1200 LT (Japan Standard Time) on 13 September 2000, and the CSD for the GY earthquake was calculated with reference to the trace at 1200 LT of 13 January 2001. The CSD calculation gives the relative variation in travel time:

  • equation image

where θi is the relative phase variation in the ith frequency component obtained by CSD calculation and ωi is its angular frequency. Angle brackets represent averaging over a certain frequency range. Here, the relative travel time is obtained by averaging them over all the available spectral components from 10 to 22 Hz. In averaging, each delay is weighed by the square root of the amplitude.

3. Changes in S Wave Travel Times Associated With Earthquakes in the Middle Distance

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Travel Time Monitoring With ACROSS
  5. 3. Changes in S Wave Travel Times Associated With Earthquakes in the Middle Distance
  6. 4. Discussion
  7. 5. Conclusion
  8. Appendix A:: Correction of the Source Instability
  9. Acknowledgments
  10. References

[18] We estimated temporal variation in the travel time of the ACROSS signal observed during the experiment using the cross-spectral density (CSD) method (Figure 5) [Ikuta et al., 2002]. Sudden delay and gradual recoveries in the travel times of the S wave (decrease in and recovery of velocity) were observed when the 2000 western Tottori (WT) earthquake and the 2001 Geiyo (GY) earthquake occurred (Figures 6a and 7c). Polarizations in the delay in S wave travel time were also found. The polarity of the travel time delay in the S wave can be attributed to an increase in the number of cracks in a certain direction. The direction that was commonly observed for the two earthquakes was parallel to the fault strike.

image

Figure 5. Temporal variation in travel time in P, transverse S, and radial S waves detected by the 800-m (black line) and 1700-m borehole (gray line) sensors along with the precipitation. The vertical arrows on the calendar axis indicate the time of the western Tottori and Geiyo earthquakes. The vertical bars indicate 95% confidence intervals of the estimated travel times according to equation (2). The gap between the couples of triangles at the right-hand side shows the resolution limit of the measurements for each sensor estimated from the signal-to-noise ratio. From Figure 4 of Ikuta et al. [2002].

Download figure to PowerPoint

image

Figure 6. (a) Travel time variations in P, radial S, and transverse S at the bottom of the 800-m borehole (solid line) and the 1700-m borehole (gray line) for 2 months including the 2000 western Tottori earthquake. The vertical gray line marks the time of the western Tottori earthquake. A rise means a delay in travel time. The bottom shows precipitation. (b) Temporal variation in phase delay near the source, shown by time delay averaged over the observed frequencies. The delays in the two horizontal directions are estimated assuming a rigid body motion of the source. (c) Horizontal strain changes at the bottom of the 800-m borehole. A rise means an extension. (d) Water level at the 800-m borehole measured by the pressure gauge installed at the top of the borehole. A rise means an increase in pressure.

Download figure to PowerPoint

image

Figure 7. (a) Travel time variations in P, radial S, and transverse S waves at the bottom of the 800-m borehole (solid line) and the 1700-m borehole (gray line) for 2 months including the 2001 Geiyo earthquake. The vertical gray line marks the time of the Geiyo earthquake. A rise means a delay in travel time. The bottom shows the precipitation hour. (b) Temporal variation in phase delay near the source, shown by time delay averaged over observed frequencies. The delays in two horizontal directions are estimated assuming the rigid body motion of the source. (c) Corrected travel time variation in the signal at the 800-m borehole sensor by referring to the vibrations near the sources. (d) Horizontal strain changes at the bottom of the 800-m borehole. A rise means extension. (e) Water level at the 800-m borehole measured by the pressure gauge installed at the bottom of the borehole. A rise means an increase in pressure.

Download figure to PowerPoint

3.1. Changes in Travel Time

[19] Figures 6a and 7a show the results of the travel time analysis applied to the P and S phases and to precipitation. The travel time of the S waves was delayed due to rainfall as well as at the time of the WT and the GY earthquakes. Gradual delays and subsequent recoveries can be seen following rainfall. The travel time delays associated with rainfall have already been reported by Yamaoka et al. [2001] and Ikuta et al. [2002]. This phenomenon is probably not due to variation in seismic wave velocity in deeper parts but due to variation in the elastic nature of the surface rock surrounding the source, which affects the motion of the source. Ikuta et al. [2002] suggest that phase variation in the source vibration is governed by the rigidity of the surrounding rock and is generally larger than that on the propagation path. Yamaoka et al. [2001] estimated variation in the signal at borehole sensors caused by variation in source motion by referring to the records of seismometers deployed near the sources. We adapted their method (see Appendix A) to the record for the GY earthquake because the signal corresponding to the GY earthquake is smaller than the WT earthquake and is greatly disturbed by temporal variation in source vibration. The correction strongly reduces the source effect shown in Figure 7c.

[20] The travel times of the S wave between the sources and the 800-m borehole sensor was suddenly delayed (the velocities decreased) by 0.4% at the time of the WT earthquake and 0.1% at the GY earthquake, although the changes were hardly observed in the P phases for either earthquake (see Figures 6a and 7c).

[21] In either case, the recovery in travel time followed the exponential decay function. The sudden delays and decay patterns suggest sudden increases and diffusions in fluid pressure associated with earthquakes. The amount of delay in the S wave travel time at the 800-m borehole sensor at the time of the WT earthquake was equivalent to that at the 1700-m borehole sensor (see Figure 6a), although the delay at the 1700-m borehole did not have sufficient resolution for the GY earthquake. The cause of the change, at least for the WT earthquake, was restricted in the areas that were shallower than 800 m.

3.2. Change in the Polarized Anisotropy of Travel Time in the S Wave

[22] The sudden delays and the recoveries of the S wave travel times showed clear polarization. We determined the direction that showed maximum delay using the following procedure. Changes in the travel time for the S waves were estimated in the directions perpendicular to the ray path using the three-component data. This analysis was applied only to the record of the 800-m borehole sensor because the record of the 1700-m borehole sensor had a smaller signal-to-noise ratio and did not have sufficient resolution to judge the direction dependence.

[23] We synthesized S waves in six directions with an interval of 30° perpendicular to the ray path from the source. Each direction was synthesized by a linear combination of the three-component data of the 800-m borehole sensor on the assumption that the ray path was on a straight line between the ACROSS sources and the sensor. Next, travel time variations in the S waves in all the directions were estimated using CSD as described in section 3.1. Each variation in the travel time for the six directions of the S wave was respectively fitted with exponential decay function A exp (τ − t)/T, as a function of time τ, in which A is the amplitude of the travel time delay, t is the time of the earthquake and T is a time constant assumed to be the same for all directions (see Figure 8).

image

Figure 8. Decay curves fit with exponential decay functions. (a) Travel time variation for 11 days around the WT earthquake fitted with exponential decay functions. The decay time constant was determined to be 2.5 days, which was assumed to be common among the six directions. The arrow marks the time of the WT earthquake (Mw 6.6). (b) Travel time variation around the GY earthquake fitted with exponential decay functions. The decay time constant was determined to be 1.5 days. Arrow marks the time of the GY earthquake (Mw 6.4).

Download figure to PowerPoint

[24] As a result, the time constant of recovery T was 2.5 days for the WT earthquake and 1.5 days for the GY earthquake. Maximum delay A was 2 ms for the WT earthquake and 0.5 ms for the GY earthquake. Figure 9 shows the amplitudes of delay A as a function of the polarization angle, in which the maximum delay was commonly observed in the N105°E direction for both the earthquakes. The direction is almost perpendicular to the fault strike of the Nojima fault.

image

Figure 9. (a) Estimated A value in the horizontal six directions at the time of the WT earthquake. The distances of the stars from the origin represent the delays in the S wave, which are fitted with an ellipse. The long axes of the ellipse were oriented in the direction N100°E. (b) A value in the horizontal six directions at the time of the GY earthquake. The long axes of the ellipse were orientated in the direction N100°E.

Download figure to PowerPoint

3.3. Static Anisotropy

[25] In sections 3.1 and 3.2, temporal variation in the travel time was estimated by calculating the cross-spectral density between the reference and target wave traces, showing there are temporal variations in anisotropy. In this section, we conducted another experiment to check whether there is static anisotropy.

3.3.1. Measurement for Existing Anisotropy Using ACROSS

[26] We again used the ACROSS source to measure the static anisotropy by splitting analysis. The splitting of S waves is observed only when the incident S wave polarizes into a certain direction. We generated such an S wave using the ACROSS sources as follows. We alternately operated each ACROSS source with normal and reverse rotations. By summing the source motions generated by the two rotations with the appropriate phase shift, we can virtually produce linear motion along the horizontal axis, which is the axis of symmetry between the rotations. The force vector generated by each rotation can be decomposed into two orthogonal components, which are parallel and normal to the axis. The parallel components are doubled, and the normal components are cancelled by summation. The same procedure can be performed for the signal recorded by the 800-m borehole sensor. We added the data that corresponds to the normal and reverse rotation after the appropriate phase shift to obtain the response for the virtual linear motions of the vibrator.

3.3.2. Polarization Direction and Travel Time Difference in Splitting S Waves

[27] To analyze S wave splitting, we applied a cross-correlation method [e.g., Shih and Meyer, 1990]. This method takes advantage of the two quasi-S phases, leading and lagging, having similarities in the leading/lagging coordinate system. The S wave splitting is characterized by two parameters, the delay time (DT) and the leading S wave polarized direction (LSPD). DT is the time difference between two quasi-S waves, which have been split in an anisotropic medium. LSPD is the polarized direction of leading quasi-S wave, which can represent the direction in which cracks along the ray path are preferentially orientated. The leading and the lagging quasi-S phases must have the highest correlation coefficient with the time shift of DT. Figure 10a shows the responses with the virtual linear vibrators that move in the direction of N50°E–N230°E, recorded using the 800-m borehole sensor. The two orthogonal components of the seismograms were synthesized from the N–S and E–W components, which were rotated clockwise from north (0°) to south (180°) by 5°. For each step, the S phases in the rotated seismograms were extracted by a time window of 0.4 s in width. The cross-correlation coefficient r between the two extracted S phases was then computed for lag time from 0 to 60 ms with increments of 2 ms. We sought the LSPD and DT that give the maximum r.

image

Figure 10. (a) Transfer functions in the horizontal component with the ACROSS signal recorded at the 800-m borehole sensor. The wave in the rectangle shows some of the S waves. The S wave in the N–S component seems to arrive faster than that in the E–W component. (b) Enlarged S wave for the rectangle in Figure 10a. (c) Particle motion for the wave trace indicated in Figure 10b. The particle motion initially polarizes in ∼5°, which is the computed LSPD, and changes abruptly at the arrival time of the lagging S wave. (d) Results of cross-correlation computation for the waveforms shown in Figure 10b. The absolute values of the cross-correlation coefficient are shown in the contour plot. The cross-correlation value is maximum at a lag time of 52 ms at a rotation angle of 5°. (e) Horizontal component seismograms that have been rotated to the leading/lagging coordinate system defined by the computed LSPD (N5°E). The lagging wave shifted forward by the computed DT (52 ms). The two quasi-S phases in the seismograms have similar shapes. (f) Particle motion of Figure 10e. This “corrected” particle motion is more linear than the original in Figure 10c. This represents the particle motion of the S wave before splitting.

Download figure to PowerPoint

[28] Figure 10d shows the cross-correlation computation for the waveform in Figure 10a. Maximum cross-correlation coefficient r of 0.99 was seen at a rotated angle of N10°E and a lag time of 50 ms, which are LSPD and DT, respectively. The LSPD coincides with the fault strike, which is identical to the direction of minimum delay at the WT and the GY earthquakes.

4. Discussion

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Travel Time Monitoring With ACROSS
  5. 3. Changes in S Wave Travel Times Associated With Earthquakes in the Middle Distance
  6. 4. Discussion
  7. 5. Conclusion
  8. Appendix A:: Correction of the Source Instability
  9. Acknowledgments
  10. References

4.1. Cause of Temporal Variation in Travel Time for the S Wave

[29] We have described changes in anisotropy associated with earthquakes in the middle distance, which appeared in the travel time for the S wave. This chapter primarily discusses their causes in terms of the cracks and fluid in them.

4.1.1. Sudden Delay in Travel Time Only for the S Wave

[30] What reduces S wave velocity with a small change in P wave velocity? It is well known that the seismic velocity is affected by the density of the cracks in the rock. O'Connell and Budiansky [1974] theoretically showed the relation between the cracked, solid, and propagation properties of seismic waves. In this relationship, seismic velocity decreases with increasing crack density, whereas the VP/VS decreases for dry cracks and increases for fluid-saturated cracks with increasing crack density. It should be noted that S wave velocity decreases at least 2.5 times more than P wave velocity when all the cracks are saturated with water. This mechanism accounts for our experimental results. The sudden delay and the following gradual recovery in the travel time that is prominent in the S wave strongly suggest that they are caused by a sudden increase and a subsequent gradual decrease in the density of the cracks that have been saturated with water.

4.1.2. Cause of Polarization in the Delay for the S Wave

[31] The polarization in the delay in the S wave suggests that cracks that are oriented in a certain direction are preferentially increased or inflated. The direction in which the S wave showed maximum delay was perpendicular to the Nojima fault and its subfault. This indicates that the density of cracks aligned in the direction parallel to the faults should be increased. According to the splitting theory of the S wave, if the incident angle is parallel to the crack planes, a lagging S wave polarizes in a direction perpendicular to the cracks. It is presumed that the cracks around this area are vertically aligned because strike-slip faults prevail in this region. Therefore the cracks should be preferentially orientated in the direction parallel to the Nojima fault and its subfault.

4.1.3. Static Crack Orientation Revealed in the Splitting Experiment

[32] We checked the existence of static anisotropy as well as temporal changes in it using S wave splitting. The polarization direction of the leading S wave LSPD and the travel time difference between the two quasi-S waves DT showed that the cracks in the LSPD were originally dominant. We therefore concluded that the cracks in the direction parallel to the Nojima fault, which were preferentially oriented, increased or inflated.

[33] The LSPD in which the cracks were oriented was N15°E, which coincides with the fault strike. This shows that the anisotropy changed in a manner that enhanced the preexisting static anisotropy. Existing cracks, which were preferentially orientated in the direction parallel to the faults, could have been suddenly opened by the increase in pore pressure and gradually closed by pressure decrease due to fluid diffusion. This shows that the cracks prevailed for even more than 5 years after the Kobe earthquake. This observation is inconsistent with the results of Tadokoro and Ando [2002].

[34] The DT, which indicates the amount of anisotropy, was about 50 ms. This was about 10% of the absolute travel time between the ACROSS sources and the 800-m borehole sensor, which seems to be fairly large for the value in the crust. Tadokoro et al. [1999] studied static anisotropy around this area using S wave splitting with natural earthquakes with focal depths of about 10 km. According to them, DTs are distributed in a range between 30 and 100 ms, which corresponds to a range between 1% and 4% of the absolute travel times. The fact that the DTs are comparable to each other suggests that the anisotropy in the upper crust is attributed to the shallowest part for this site.

4.2. Strain and Water Pressure

[35] Temporal changes in the travel time and splitting of the S wave showed that the inflation of pore pressure in the preferentially oriented cracks enhanced anisotropy during the WT and the GY earthquakes. We show that these interpretations explain other observations, such as the change in strain and pore pressure measured in the borehole.

4.2.1. Areal Strain

[36] Horizontal strain was monitored at the bottom of the 800-m borehole with a three-component strain meter. Figures 6c and 7d show temporal variation for the horizontal strain during the WT and the GY earthquakes. In all directions, we can find compression at certain rate regarded to be constant for a few months. The strain showed unusual change when both of the earthquakes occurred. In the WT earthquake, all the components showed an instantaneous step at the moment the earthquake took place, subsequently followed by rapid compression and gradual recovery. Compression after the GY earthquake and subsequent recovery without instantaneous steps were observed. Ignoring the instantaneous steps and subtracting the constant trend, the amount of compressive strain rose to 4 × 10−7 and 3 × 10−8 in the case of the WT and the GY earthquakes, respectively.

[37] We calculated changes in the strain due to slips in the WT and the GY earthquakes using simulation software for elastic deformation MICAP-G [Naito and Yoshikawa, 1999; Okada, 1992] to confirm whether the observed strains can be explained by the elastic response to fault dislocations at the sources. In the case of the WT earthquake, the areal strain around the experiment site should be dilatant, and the amplitude should be at most 10−8, whereas the observed strain was no less than 5 × 10−7 in compression. In the case of the GY earthquake, the areal strain around the experiment site should be contracted, and the amplitude should be at most 5 × 10−9, whereas the observed strain was no less than 10−7. The observed strains were therefore inconsistent with the elastic response to fault dislocations at the sources. The peaks of the observed compressions for both cases were delayed for about 2 days from the occurrence of the WT and the GY earthquakes, although the delay in the travel time for the S wave peaked immediately after the earthquake. This can be explained by pore pressure diffusion, which was also described by the decay pattern of the travel time delay. Pore pressure, which increased in the shallower portion at the time of each earthquake, should have diffused downward for a few days.

4.2.2. Maximum Compressive Strain Axis

[38] Anisotropy for the strain changes was also remarkable, and this is discussed in this section. However, we must first mention the misorientation of the strain meter. The strain meter is built in a multicomponent borehole instrument. The instrument consists of a velocity sensor (the 800-m borehole sensor), a strain meter, a tiltmeter, and an accelerometer [Ishii et al., 2002]. We checked the azimuthal direction of the velocity sensor and the accelerometer using initial motion of the P waves of nearby earthquakes, and revealed that both of them rotated clockwise from the originally intended orientation. The angles were 84° and 88°, respectively, with an error of plus or minus 7°. We can consider that the whole system of multicomponent instruments rotates together. We discuss the records of the strain meter assuming that it was rotated 88°.

[39] Anisotropy for the strain changes in the two cases was similar. Figures 11a and 11b show the principal strain axes with the peak amplitudes following the WT and the GY earthquakes, respectively. The directions of the maximum principal strain were oriented to N110°E, perpendicular to the fault strike for both cases. It can be explained by the preferred orientation of the cracks, which was revealed by the polarized anisotropy of travel time. The polarization anisotropy of the travel time shows that most of the existing cracks were aligned parallel to the fault strike, for which the integrated internal pressure over the broad side of the cracks could have selectively compressed the strain meter in the direction (see Figure 12) of maximum compressive strain.

image

Figure 11. Horizontal compressive strains observed by the three-component strain meter deployed in the 800-m borehole at the time of (a) the WT earthquake and (b) the GY earthquake. The distance of the stars from the origin represents the observed compressive strains. The two orthogonal solid lines show the principle strain axes.

Download figure to PowerPoint

image

Figure 12. Matrix granite (gray region). The shaded ellipses show fluid-saturated cracks. If preferentially orientated cracks (larger cracks) are opened by pore pressure inflation, the compressive stress in the direction normal to the cracks that predominate in this medium and the strain meter is compressed in this direction. The broad side (shown by arrows) selectively compresses the strain meter.

Download figure to PowerPoint

4.2.3. Pore Pressure

[40] Water pressure at the bottom of the 800-m borehole had also been monitored (see Figures 6d and 7e). The water in the borehole is supplied through the strainer holes at the bottom, and the pressure is measured at the top where the hole is sealed. We may assume that the change in the measured pressure is identical to that in the aquifer that is connected to the strainers at the bottom. The water pressure increased beyond the range of measurement at the time of the WT earthquake, which was at least 100 kPa, followed by exponential decay for a few days. It is also consistent with the assumption of the inflation and diffusion of pore pressure. In contrast to the large increase in water pressure for the WT, no increase or decrease was detected at the time of the GY earthquake. Considering pore pressure inflation, one likely explanation for the lack of response is that permeability around the strainer was significantly reduced after the WT earthquake. If the strainer was clogged, the water would have been unable to permeate the borehole. In this case, the pore pressure in the surrounding crack may have affected the strain meter without any change in the water pressure in the well.

4.3. Relation Between Travel Time Changes and Earthquake Ground Motion

[41] Four earthquakes with relatively large ground motion were observed at the experiment site during our experiment. Of these, only the WT and the GY earthquakes caused a delay in the travel times of the S waves. This leads us to consider the cause of the pore pressure increase. The magnitudes of the four earthquakes were Mw 6.6 (WT), Mw 6.4 (GY), Mw 5.4 and Mw 4.2, with epicenter distances of 165 km, 215 km, 110 km and 85 km from our site, respectively. The ground motion of the four earthquakes was recorded by strong motion accelerometers deployed around the site (Figure 1). The maximum amplitude of acceleration at the time of each earthquake is shown in Figure 13b. The acceleration due to the WT earthquake was about 1.5 times as large as that due to the GY earthquake, which was about twice as large as that due to the other two. The delays in travel time at the time of the WT earthquake was about 4 times as large as that of the GY earthquake, although no delay was detected at the time of the other two earthquakes.

image

Figure 13. (a) Wave traces at the time of the four earthquakes recorded by the E–W component of an accelerometer deployed at a station by NIED (see Figure 1). The station was 15 km from the experiment site. From top to bottom, waveforms of the WT and the GY earthquakes, and the other two earthquakes, are shown. (b) Maximum acceleration of the four earthquakes observed in Awaji Island. Each bar indicates the maximum observed acceleration averaged over four accelerometers deployed on Awaji Island.

Download figure to PowerPoint

[42] Although the change in travel time and the ground motion positively correlate, clear nonlinearity exists between them. This suggests that there are certain thresholds in the amplitude of shaking that cause changes in the travel time. The threshold level may indicate a kind of strength that can be exceeded by violent shaking. For example, when highly pressurized water in an isolated volume in a deeper portion moves into the cracks in the shallower part, the movement of water can be triggered by opening water channels due to earthquake shaking. The material, which has sealed the water channels, is broken by the shaking of the earthquakes. In the WT and the GY earthquakes, the high pressure propagated through the channels from the pressure sources to the region where the ray path of the ACROSS signal runs. In the case of the WT earthquake, the strong shaking broke more of the seals in the channels between the pressure source and the ray path than in the case of the GY earthquake, and a larger amount of pressurized water moved. Thus both delay amplitude A and diffusing time constant T should be greater for strong shaking. In the other two earthquakes, shaking did not exceed the strength of the seals.

4.4. Implications for Groundwater Movement

[43] Many papers reported changes in seismicities, strain, and groundwater discharge associated with earthquakes in middle distances [Hill et al., 1993, 1995; Rojstaczer et al., 1995; Kitagawa and Koizumi, 2000; Ogasawara et al., 2002]. For example, Hill et al. [1993] reported that the 1992 Landers earthquake triggered a sudden and widespread increase in earthquake activity across much of the western United States. They suspected crustal fluids as a cause of this phenomenon based on the temporal patterns of triggered activity. Kitagawa and Koizumi [2000] repeatedly detected unusual changes in discharge rate and water temperature at Yutani hot spring, Japan, in response to nearby or distant earthquakes. They mentioned a possibility that a local strain release, such as aseismic slip at a fault near the site, might have caused the large strain changes which were estimated from the coseismic groundwater change. Ogasawara et al. [2002] observed a significant change in strain, self-potential, and microseismic activity associated with the 1995 Kobe earthquake at Ikuno mine, which is located at 67 km northwest of the epicenter. They concluded that the change was due to an increase of pore pressure by earthquake shaking. In all the papers above, they attributed the change to the change in the pore pressure of crustal fluid. They, however, only gave speculative idea for the pressure change. In this study, we are able to propose a mechanism for pore pressure increase with a new observation both in quantitatively and qualitatively.

5. Conclusion

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Travel Time Monitoring With ACROSS
  5. 3. Changes in S Wave Travel Times Associated With Earthquakes in the Middle Distance
  6. 4. Discussion
  7. 5. Conclusion
  8. Appendix A:: Correction of the Source Instability
  9. Acknowledgments
  10. References

[44] Temporal variation in seismic velocity near the Nojima Fault, which ruptured during the 1995 Kobe earthquake (Mw 7.2), was monitored using an accurately controlled routine-operated signal system (ACROSS). Temporal change in P and S between the ACROSS source at the surface and a seismometer in the 800-m borehole were monitored. Sudden decreases in the velocity of the S waves were observed when the 2000 western Tottori earthquake (Mw 6.6) and the 2001 Geiyo earthquake (Mw 6.4) occurred. Their epicenters were 165 and 215 km away from the experiment site, respectively. The travel time of the S wave was abruptly delayed at the time of each earthquake, and there was gradual recovery to the original travel time, though few changes were observed in the P waves. The delay in S was about 0.4% and 0.1% for the western Tottori earthquake and the Geiyo earthquake, respectively. According to the analysis for anisotropy, the delays were polarized in a direction perpendicular to the Nojima fault. Analysis for the static anisotropy using ACROSS also shows that there is strong anisotropy in which LSPD oriented parallel to the Nojima fault. The change in the anisotropy occurred in such a manner that the existing anisotropy, which were originally dominant in the direction probably due to fault fracturing, was enhanced. The most likely explanation for the anisotropic change is that the pore pressure of the groundwater in the shallow portion increased due to water channels opening because of the shaking from the earthquakes in the middle distance. These results explain the anomalous changes in the strain observations.

Appendix A:: Correction of the Source Instability

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Travel Time Monitoring With ACROSS
  5. 3. Changes in S Wave Travel Times Associated With Earthquakes in the Middle Distance
  6. 4. Discussion
  7. 5. Conclusion
  8. Appendix A:: Correction of the Source Instability
  9. Acknowledgments
  10. References

[45] Variation in the travel time of the ACROSS signal observed at the borehole sensors is strongly affected by variation in the motion of the sources. We corrected the variation in the ACROSS signal received by the 800-m borehole sensor with reference to the sensors deployed at the ground surface. We assumed the foundation to which the ACROSS sources are attached to be the signal source. We also assumed that the wave field at the borehole sensors is modeled with a linear combination of the signals at the seismometers placed near the sources (Figure A1). We can therefore express the observed vibrations at sensor Y(ω) (Figure A2)with the product of transfer function G(ω) and the vibration near sources X(ω). In our observation, both G(ω) and X(ω) are assumed to vary with time. The observed vibrations are represented by

  • equation image

where ω is the angular frequency used as the ACROSS signal, and t is the time. G0 and X0 denote the time-invariant parts of G and X, respectively. We assumed that the term in which there are variations with time was short and denoted it as Δ. Ignoring higher-order fractions, it can be rewritten as

  • equation image

We obtained G0 using the following procedure. As the motion of the foundation is described by the output of the sensors with 12 components in total, we can write the vibrations at the borehole sensors with a linear combination of all the components and the residual:

  • equation image

in which R is the residual, i denotes each component of the vibrations of the borehole sensors, j that of the sensors near the sources, and, k, the serial number of the samples in the observation period. The summation convention is employed in the equations given hereafter. Note that Gij0 is independent of time. Gij0 is obtained using least squares fitting to the observed data of the borehole sensors. Figure A3a shows the predicted temporal variation

  • equation image

using the transfer function obtained here, for example, for 14 Hz. Comparing temporal variation in the predicted motion with the observed motion (Figure A2), we can see that they are quite similar, meaning that the temporal variations observed at the seismometer is attributed mostly to variations in the motion of the foundation.

image

Figure A1. Temporal variation in the amplitude and phase of the 14-Hz signal measured at the four seismometers placed on the foundation and in the 10-m-deep borehole (see Figure 2).

Download figure to PowerPoint

image

Figure A2. Temporal variation in the amplitude and phase signal at 14 Hz received by the 800-m borehole sensor. The variation pattern of the amplitude is similar to that for the signal measured by the sensors in the vibrator house (see Figure 2). Phase delay is observed at the time of the Geiyo earthquake (marked by an arrow).

Download figure to PowerPoint

image

Figure A3. (a) Predicted variation in the signal at the 800-m borehole sensor. These data are produced with transfer function Gji0 obtained here and the motion of the ground near the sources (Figure A1). The variation patterns are very similar to those of the observed patterns (see Figure A2). (b) Corrected temporal variation in the signal of the borehole sensor. Note that the variations are reduced, and the change of phase at the time of the Geiyo earthquake becomes clear.

Download figure to PowerPoint

[46] In this calculation, temporal variation in the medium (i.e., ΔG(ω, t)) is contained in the residual Ri(k). Following equation (A2), the residual is written as

  • equation image

Here, we want to obtain the vibrations of the seismometer after correction of the temporal variation in the foundation:

  • equation image

From equation (A1), we may assume

  • equation image

in which angle brackets denote the average over time k. This is equivalent to the assumption that

  • equation image

That is, time-invariant term X is defined so that the temporal average of ΔX vanishes.

[47] Figure A3b shows 〈Yi(k)〉 + Ri(k) at 14 Hz against time. In Figure A3b, temporal variation, especially diurnal and long-term variation, is dramatically reduced. The amplitude variation lowers to less than 10%, and diurnal and long-term phase variation mostly disappears. In the phase variation, the change that occurred during the Geiyo earthquake remains very clear. In our analysis, we consider 〈Yi(k)〉 + Ri(k) as the record at the boreholes, which is free from the effect of variation near the sources. The results of the travel time analysis using 〈Yi(k)〉 + Ri(k) is shown in Figure 7c.

Acknowledgments

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Travel Time Monitoring With ACROSS
  5. 3. Changes in S Wave Travel Times Associated With Earthquakes in the Middle Distance
  6. 4. Discussion
  7. 5. Conclusion
  8. Appendix A:: Correction of the Source Instability
  9. Acknowledgments
  10. References

[48] A number of people have generously cooperated in this study. In particular, we would like to thank T. Kunitomo, R. Miyajima, T. Okuda, A. Saiga, H. Misu, and K. Tsuruga in the field. We thank K. Fujimori, A. Mukai, and H. Ito for their permission to use unpublished data. We specially thank M. Kumazawa, K. Tadokoro, T. Watanabe, and N. Fujii for their critical comments on the manuscript.

References

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Travel Time Monitoring With ACROSS
  5. 3. Changes in S Wave Travel Times Associated With Earthquakes in the Middle Distance
  6. 4. Discussion
  7. 5. Conclusion
  8. Appendix A:: Correction of the Source Instability
  9. Acknowledgments
  10. References
  • Aggarwal, Y. P., D. W. Simpson, and L. R. Sykes (1974), Temporal and spatial analysis of premonitory velocity anomalies for the Aug. 3, 1973 Blue Mt. Lake (BML) earthquake, Eos Trans. AGU, 55(4), 355.
  • Aster, R., P. Shearer, and J. Berger (1990), Quantitative measurements of shear wave polarizations at the Anza seismic network, southern California: Implications for shear wave splitting and earthquake prediction, J. Geophys. Res., 95, 12,44912,473.
  • Bokelmann, G. H. R., and H. P. Harjes (2000), Evidence for temporal variation of seismic velocity within the upper continental crust, J. Geophys. Res., 105, 23,87923,894.
  • Crampin, S. (1978), Seismic-wave propagation through a cracked solid: Polarization as a possible dilatancy diagnostic, Geophys. J. R. Astron. Soc., 53(3), 467496.
  • Crampin, S. (1987), Geological and industrial implications of extensive-dilatancy anisotropy, Nature, 328, 491496.
  • Hill, D. P., et al. (1993), Seismicity remotely triggered by the magnitude 7.3 Landers, California, earthquake, Science, 260, 16171623.
  • Hill, D. P., M. J. S. Johnston, J. O. Langbein, and R. Bilham (1995), Response of Long Valley caldera to the Mw = 7.3 Landers, California, earthquake, J. Geophys. Res., 100, 12,98513,005.
  • Ikuta, R., K. Yamaoka, K. Miyakawa, T. Kunitomo, and M. Kumazawa (2002), Continuous monitoring of propagation velocity of seismic wave using ACROSS, Geophys. Res. Lett., 29(13), 1627, doi:10.1029/2001GL013974.
  • Ishii, H., T. Yamauchi, S. Matsumoto, Y. Hirata, and S. Nakao (2002), Development of multi-component borehole instrument for earthquake prediction study: Some observed examples of precursory and co-seismic phenomena relating to earthquake swarms and application of the instrument for rock mechanics, in Seismogenic Process Monitoring: Proceedings of a Joint Japan-Poland Symposium on Mining and Experimental Seismology, Kyoto, Japan, November 1999, edited by H. Ogasawara, T. Yanagidani, and M. Ando, pp. 365377, A. A. Balkema, Brookfield, Vt.
  • Karageorgi, E., R. Clymer, and T. V. McEvilly (1992), Seismological studies at Parkfield; II, Search for temporal variations in wave propagation using vibroseis, Bull. Seismol. Soc. Am., 82, 13881415.
  • Kitagawa, Y., and N. Koizumi (2000), A study on the mechanism of coseismic groundwater changes: Interpretation by a groundwater model composed of multiple aquifers with different strain responses, J. Geophys. Res., 105, 19,12119,134.
  • McEvilly, T. V., and L. R. Johnson (1974), Stability of P and S velocities from central California quarry blasts, Bull. Seismol. Soc. Am., 64, 343353.
  • Meteorological Research Institute (1999), Coseismic strain step observed by an Ishii-type 3-component strainmeter at Imazu, generated by an earthquake in northern Shiga Prefecture in March, Rep. 62, pp. 428431, Coord. Comm. for Earthquake Predict., Tsukuba, Japan.
  • Naito, H., and S. Yoshikawa (1999), A program to assist crustal deformation analysis (in Japanese), J. Seismol. Soc. Jpn., 52, 101103.
  • O'Connell, R. J., and B. Budiansky (1974), Seismic velocities in dry and saturated cracked solids, J. Geophys. Res., 79, 54125426.
  • Ogasawara, H., Y. Kuwabara, T. Miwa, K. Fujimori, N. Hirano, and M. Koizumi (2002), Post-seismic effects of an M 7.2 earthquake and micro seismicity in an abandoned, flooded, deep mine, Pure Appl. Geophys., 159, 6390.
  • Okada, Y. (1992), Internal deformation due to shear and tensile faults in a half-space, Bull. Seismol. Soc. Am., 82, 10181040.
  • Reasenberg, P., and K. Aki (1974), A precise, continuous measurement of seismic velocity for monitoring in situ stress, J. Geophys. Res., 79, 399406.
  • Rojstaczer, S., S. Wolf, and R. Michel (1995), Permeability enhancement in the shallow crust as a cause of earthquake-induced hydrological changes, Nature, 373, 237239.
  • Saiga, A., Y. Hiramatsu, T. Ooida, and K. Yamaoka (2003), Spatial variation in the crustal anisotropy and its temporal variation associated with a moderate-sized earthquake in the Tokai region, central Japan, Geophys. J. Int., 154(3), 695705.
  • Semenov, A. N. (1969), Variations in the travel-time of transverse and longitudinal waves before violent earthquakes, Izv. Akad. Nauk SSSR, Fiz. Zemli., 4, 7277.
  • Shih, X., and R. Meyer (1990), Observation of shear wave splitting from natural events: South moat of Long Valley caldera, California, June 29 to August 12, 1982, J. Geophys. Res., 95, 11,17911,195.
  • Tadokoro, K., and M. Ando (2002), Evidence for rapid fault healing derived from temporal changes in S wave splitting, Geophys. Res. Lett., 29(4), 1047, doi:10.1029/2001GL013644.
  • Tadokoro, K., M. Ando, and Y. Umeda (1999), S wave splitting in the aftershock region of the 1995 Hyogo-ken Nanbu earthquake, J. Geophys. Res., 104, 981991.
  • Terashima, T. (1974), Change of VP/VS before the large earthquake of April 1, 1968 in Hyuganada, Japan, Bull. Int. Inst. Seismol. Earth. Eng., 12, 1729.
  • Yamaoka, K., T. Kunitomo, K. Miyakawa, K. Kobayashi, and M. Kumazawa (2001), A trial for monitoring temporal variation of seismic velocity with ACROSS system, Island Arc, 10, 336347.
  • Zatsepin, S. V., and S. Crampin (1997), Modeling the compliance of crustal rock: I-Response of shear-wave splitting to differential stress, Geophys. J. Int., 129, 477494.